checkout bugfix: use stat.mtime instead of stat.ctime in two places
Commit e1afca4fd "write_index(): update index_state->timestamp after
flushing to disk" on 2009-02-23 used stat.ctime to record the
timestamp of the index-file. This is wrong, so fix this and use the
correct stat.mtime timestamp instead.
Commit 110c46a909 "Not all systems use st_[cm]tim field for ns
resolution file timestamp" on 2009-03-08, has a similar bug for the
builtin-fetch-pack.c file.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The comments indicated that setting a Makefile variable USE_NSEC would
enable the code for sub-second [cm]times. However, the Makefile
variable was never turned into a compiler switch so the code was never
enabled. This patch allows USE_NSEC to be noticed by the compiler.
Signed-off-by: Brian Gernhardt <benji@silverinsanity.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Not all OSes use st_ctim and st_mtim in their struct stat. In
particular, it appears that OS X uses st_*timespec instead. So add a
Makefile variable and #define called USE_ST_TIMESPEC to switch the
USE_NSEC defines to use st_*timespec.
This also turns it on by default for OS X (Darwin) machines. Likely
this is a sane default for other BSD kernels as well, but I don't have
any to test that assumption on.
Signed-off-by: Brian Gernhardt <benji@silverinsanity.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Not all systems use st_[cm]tim field for ns resolution file timestamp
Some codepaths do not still use the ST_[CM]TIME_NSEC() pair of macros
introduced by the previous commit but assumes all systems use st_mtim
and st_ctim fields in "struct stat" to record nanosecond resolution part
of the file timestamps.
Record ns-timestamps if possible, but do not use it without USE_NSEC
Traditionally, the lack of USE_NSEC meant "do not record nor use the
nanosecond resolution part of the file timestamps". To avoid problems on
filesystems that lose the ns part when the metadata is flushed to the disk
and then later read back in, disabling USE_NSEC has been a good idea in
general.
If you are on a filesystem without such an issue, it does not hurt to read
and store them in the cached stat data in the index entries even if your
git is compiled without USE_NSEC. The index left with such a version of
git can be read by git compiled with USE_NSEC and it can make use of the
nanosecond part to optimize the check to see if the path on the filesystem
hsa been modified since we last looked at.
write_index(): update index_state->timestamp after flushing to disk
Since this timestamp is used to check for racy-clean files, it is
important to keep it uptodate.
For the 'git checkout' command without the '-q' option, this make a
huge difference. Before, each and every file which was updated, was
racy-clean after the call to unpack_trees() and write_index() but
before the GIT process ended.
And because of the call to show_local_changes() in builtin-checkout.c,
we ended up reading those files back into memory, doing a SHA1 to
check if the files was really different from the index. And, of
course, no file was different.
With this fix, 'git checkout' without the '-q' option should now be
almost as fast as with the '-q' option, but not quite, as we still do
some few lstat(2) calls more without the '-q' option.
Below is some average numbers for 10 checkout's to v2.6.27 and 10 to
v2.6.25 of the Linux kernel, to show the difference:
before (git version 1.6.2.rc1.256.g58a87):
7.860 user 2.427 sys 19.465 real 52.8% CPU faults: 0 major 95331 minor
after:
6.184 user 2.160 sys 17.619 real 47.4% CPU faults: 0 major 38994 minor
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Signed-off-by: Junio C Hamano <gitster@pobox.com>
If we inside verify_uptodate() can already tell from the ce entry that
it is already uptodate by testing it with ce_uptodate(ce), there is no
need to call lstat(2) and ie_match_stat() afterwards.
Since the filesystem ext4 is now defined as stable in Linux v2.6.28,
and ext4 supports nanonsecond resolution timestamps natively, it is
time to make USE_NSEC work as expected.
This will make racy git situations less likely to happen. For 'git
checkout' this means it will be less likely that we have to open, read
the contents of the file into RAM, and check if file is really
modified or not. The result sould be a litle less used CPU time, less
pagefaults and a litle faster program, at least for 'git checkout'.
Since the number of possible racy git situations would increase when
disks gets faster, this patch would be more and more helpfull as times
go by. For a fast Solid State Disk, this patch should be helpfull.
Note that, when file operations starts to take less than 1 nanosecond,
one would again start to get more racy git situations.
For more info on racy git, see Documentation/technical/racy-git.txt
For more info on ext4, see http://kernelnewbies.org/Ext4
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Signed-off-by: Junio C Hamano <gitster@pobox.com>
'struct cache' does not have a 'usec' member, but a 'unsigned int
nsec' member. Simmilar 'struct stat' does not have a 'st_mtim.usec'
member, and we should instead use 'st_mtim.tv_nsec'.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Notice the memmove line, where the CPU did 7710 / 277 = 27.8 cycles
per instruction, and compared to the total cycles spent inside the
source code of GIT for this command, all the memmove() calls
translates to (7710 * 100) / 14775 = 52.2% of this.
Retesting with a GIT program compiled for gcov usage, I found out that
the memmove() calls came from remove_index_entry_at() in read-cache.c,
where we have:
remove_index_entry_at() is called 4902 times from check_updates() in
unpack-trees.c, and each time called we move each cache_entry pointers
(from the removed one) one step to the left.
Since we have 28828 entries in the cache this time, and if we on
average move half of them each time, we in total move approximately
4902 * 0.5 * 28828 * 4 = 282 629 712 bytes, or twice this amount if
each pointer is 8 bytes (64 bit).
OK, is seems that the function check_updates() is called 28 times, so
the estimated guess above had been more correct if check_updates() had
been called only once, but the point is: we get lots of bytes moved.
To fix this, and use an O(N) algorithm instead, where N is the number
of cache_entries, we delete/remove all entries in one loop through all
entries.
From a retest, the new remove_marked_cache_entries() from the patch
below, ended up with the following output line from oprofile:
If we can trust the numbers from oprofile in this case, we saved
approximately ((7710 - 46) * 20000) / (2 * 1000 * 1000 * 1000) = 0.077
seconds CPU time with this fix for this particular test. And notice
that now the CPU did only 46 / 15 = 3.1 cycles/instruction.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
lstat_cache(): print a warning if doing ping-pong between cache types
This is a debug patch which is only to be used while the lstat_cache()
is in the test stage, and should be removed/reverted before the final
relase.
I think it should be useful to catch these warnings, as I it could be
an indication of that the cache would not be very effective if it is
doing ping-pong by switching between different cache types too many
times.
Also, if someone is experimenting with the lstat_cache(), this patch
will maybe be useful while debugging.
If someone is able to trigger the warning, then send a mail to the GIT
mailing list, containing the first 15 lines of the warning, and a
short description of the GIT commands to trigger the warnings.
I hope someone is willing to use this patch for a while, to be able to
catch possible ping-pong's.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Currently inside show_patch_diff() we have an fstat() call after an
ok lstat() call. Since before the call to fstat() we have already
tested for the link case with S_ISLNK(), the fstat() can be removed.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Signed-off-by: Junio C Hamano <gitster@pobox.com>
write_entry(): use fstat() instead of lstat() when file is open
Currently inside write_entry() we do an lstat(path, &st) call on a
file which have just been opened inside the exact same function. It
should be better to call fstat(fd, &st) on the file while it is open,
and it should be at least as fast as the lstat() method.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Currently inside unlink_entry() if we get a successful removal of one
file with unlink(), we try to remove the leading directories each and
every time. So if one directory containing 200 files is moved to an
other location we get 199 failed calls to rmdir() and 1 successful
call.
To fix this and avoid some unnecessary calls to rmdir(), we schedule
each directory for removal and wait much longer before we do the real
call to rmdir().
Since the unlink_entry() function is called with alphabetically sorted
names, this new function end up being very effective to avoid
unnecessary calls to rmdir(). In some cases over 95% of all calls to
rmdir() is removed with this patch.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Simplify the if-else test in longest_match_lstat_cache() such that we
only have one simple if test. Instead of testing for 'i == cache.len'
or 'i == len', we transform this to a common test for 'i == max_len'.
And to further optimise we use 'i >= max_len' instead of 'i ==
max_len', the reason is that it is now the exact opposite of one part
inside the while-loop termination expression 'i < max_len && name[i]
== cache.path[i]', and then the compiler can probably reuse a test
instruction from it.
We also throw away the arguments to reset_lstat_cache(), such that all
the safeguard logic inside lstat_cache() is handled at one place.
Signed-off-by: Kjetil Barvik <barvik@broadpark.no> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The previous check in git-web--browse for /bin/start used test -n
/bin/start, which was always true. This lead to "start" being tried
first in the browser list. On systems with upstart installed, "start"
exists and might be in the PATH, but it makes a poor choice for a web
browser. Instead, test that /bin/start exists and is executable.
Signed-off-by: Todd Zullinger <tmz@pobox.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
gitweb: add $prevent_xss option to prevent XSS by repository content
Add a gitweb configuration variable $prevent_xss that disables features
to prevent content in repositories from launching cross-site scripting
(XSS) attacks in the gitweb domain. Currently, this option makes gitweb
ignore README.html (a better solution may be worked out in the future)
and serve a blob_plain file of an untrusted type with
"Content-Disposition: attachment", which tells the browser not to show
the file at its original URL.
The XSS prevention is currently off by default.
Signed-off-by: Matt McCutchen <matt@mattmccutchen.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
add -p: get rid of Git.pm warnings about unitialized values
After invoking git add -p I always got the warnings:
Use of uninitialized value $_[3] in exec at Git.pm line 1282.
Use of uninitialized value $args[2] in join or string at Git.pm line 1264.
A bisect showed that these warnings occur in a301973 "add -p: print errors
in separate color" the first time.
They can be reproduced by setting color.ui (or color.interactive) to "auto"
and unsetting color.interactive.help and color.interactive.error.
I am using Perl 5.10.0.
The reason of the warning is that color.interactive.error defaults to
color.interactive.help which defaults to nothing in the specific codepath.
It defaults to 'red bold' some lines above which could lead to the wrong
assumption that it always defaults to 'red bold' now.
This patch lets it default to 'red bold', blowing the warnings away.
Signed-off-by: Stephan Beyer <s-beyer@gmx.net> Acked-By: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
rev-list: fix showing distance when using --bisect-all
Before d467a52 ("Make '--decorate' set an explicit 'show_decorations'
flag", Nov 3 2008), commit decorations were shown whenever they exist, and
distances stored in them by "git rev-list --bisect-all" were automatically
shown. d467a52 changed the rule so that commit decorations are not shown
unless rev_info explicitly asks to, with its show_decorations bit, but
forgot that the ones "git rev-list --bisect-all" adds need to be shown.
This patch fixes this old breakage.
Signed-off-by: Christian Couder <chriscool@tuxfamily.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* js/filter-branch-submodule:
filter-branch: do not consider diverging submodules a 'dirty worktree'
filter-branch: Fix fatal error on bare repositories
emacs: Remove the no longer maintained vc-git package.
vc-git is distributed with Emacs since version 22.2, and is maintained
in the Emacs CVS tree. This file is obsolete and causes trouble for
people who want to add contrib/emacs to their load-path.
eval{use...} is no good because the 'use' is evaluated at compile
time, so manually 'require' it. We need to forward declare the
functions we use, otherwise Perl raises a compilation error.
Signed-off-by: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* js/maint-remote-remove-mirror:
builtin-remote: make rm operation safer in mirrored repository
builtin-remote: make rm() use properly named variable to hold return value
* js/notes:
git-notes: fix printing of multi-line notes
notes: fix core.notesRef documentation
Add an expensive test for git-notes
Speed up git notes lookup
Add a script to edit/inspect notes
Introduce commit notes
* rc/http-push:
http-push: wrap signature of get_remote_object_url
http-push: add back underscore separator before lock token
http-push.c: get_remote_object_url() is only used under USE_CURL_MULTI
http-push: refactor request url creation
* cb/mergetool:
mergetool: fix running mergetool in sub-directories
mergetool: Add a test for running mergetool in a sub-directory
mergetool: respect autocrlf by using checkout-index
filter-branch: do not consider diverging submodules a 'dirty worktree'
At the end of filter-branch in a non-bare repository, the work tree is
updated with "read-tree -m -u HEAD", to carry the change forward in case
the current branch was rewritten. In order to avoid losing any local
change during this step, filter-branch refuses to work when there are
local changes in the work tree.
This "read-tree -m -u HEAD" operation does not affect what commit is
checked out in a submodule (iow, it does not touch .git/HEAD in a
submodule checkout), and checking if there is any local change to the
submodule is not useful.
Staged submodules _are_ considered to be 'dirty', however, as the
"read-tree -m -u HEAD" could result in loss of staged information
otherwise.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use Term::ReadKey, if available and enabled with interactive.singlekey,
to let the user answer add -p's prompts by pressing a single key. We're
not doing the same in the main 'add -i' interface because file selection
etc. may expect several characters.
Two commands take an argument: 'g' can easily cope since it'll just
offer a choice of chunks. '/' now (unconditionally, even without
readkey) offers a chance to enter a regex if none was given.
Signed-off-by: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Makefile: minor improvements for Mac OS X (Darwin)
1) Instead of requesting OLD_ICONV on all Mac OS X versions except for 10.5
(which will break when 10.6 is released), exlicitly request it for versions
older than 10.5.
2) NO_STRLCPY is not needed since Mac OS X 10.2. Noticed by Benjamin Kramer.
Note that uname -r returns the underlying Darwin version, which can be mapped
to Mac OS X version at http://www.opensource.apple.com/darwinsource/
Signed-off-by: Jay Soffian <jaysoffian@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The installation rules wanted to differentiate between a template_dir that
is given as an absolute path (e.g. /usr/share/git-core/templates) and a
relative one (e.g. share/git-core/templates) but it was done by checking
if $(abspath $(template_dir)) and $(template_dir) yield the same string.
This was wrong in at least two ways.
* The user can give template_dir with a trailing slash from the command
line to invoke make or from the included config.mak. A directory path
ought to mean the same thing with or without such a trailing slash but
use of $(abspath) means an absolute path with a trailing slash fails
the test.
* Versions of GNU make older than 3.81 do not have $(abspath) to begin
with.
This changes the detection logic to see if the given path begins with a
slash.
After the git-core package was renamed to git, git help -w was still looking
for files in /usr/share/doc/git-core-$VERSION instead of
/usr/share/doc/git-$VERSION.
Signed-off-by: David J. Mellor <dmellor@whistlingcat.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
builtin-blame.c: Use utf8_strwidth for author's names
git blame misaligns output if a author's name has a differing display width and
strlen; for instance, an accented Latin letter that takes two bytes to encode
will cause the rest of the line to be shifted to the left by one. To fix this,
use utf8_strwidth instead of strlen (and compute the padding ourselves, since
printf doesn't know about UTF-8).
Signed-off-by: Geoffrey Thomas <geofft@mit.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This rewrites the example part of the bundle doucmentation to follow
the suggestion made by Junio during a recent discussion (gmane 108030).
Instead of just showing different ways to create and use bundles in a
disconnected fashion, the rewritten example first shows the simplest
"full cycle" of sneakernet workflow, and then introduces various
variations.
The words are mostly taken from Junio's outline. I only reformatted
them and proofread to make sure the end result flows naturally.
Signed-off-by: Nanako Shiraishi <nanako3@lavabit.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
contrib git-resurrect: find traces of a branch name and resurrect it
Add a tool 'git-resurrect.sh <branch>' that tries to find traces of
the <branch> in the HEAD reflog and, optionally, all merge commits in
the repository. It can then resurrect the branch, pointing it at the
most recent of all candidate commits found.
Signed-off-by: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint:
urls.txt: document optional port specification in git URLS
builtin-mv.c: check for unversionned files before looking at the destination.
Add a testcase for "git mv -f" on untracked files.
Missing && in t/t7001.sh.
* maint-1.6.0:
builtin-mv.c: check for unversionned files before looking at the destination.
Add a testcase for "git mv -f" on untracked files.
Missing && in t/t7001.sh.
builtin-mv.c: check for unversionned files before looking at the destination.
The previous code was failing in the case where one moves an
unversionned file to an existing destination, with mv -f: the
"existing destination" was checked first, and the error was cancelled
by the force flag.
We now check the unrecoverable error first, which fixes the bug.
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
builtin-remote: make rm operation safer in mirrored repository
"git remote rm <repo>" happily removes non-remote refs and their reflogs.
This may be okay if the repository truely is a mirror, but if the user
had done "git remote add --mirror <repo>" by accident and was just
undoing their mistake, then they are left in a situation that is
difficult to recover from.
After this commit, "git remote rm" skips over non-remote refs. The user
is advised on how remove branches using "git branch -d", which itself
has nice safety checks wrt to branch removal lacking from "git remote rm".
Non-remote non-branch refs are skipped silently.
Signed-off-by: Jay Soffian <jaysoffian@gmail.com> Acked-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
If the user hit Ctrl-D (EOF) while the script was in 'go to hunk?'
mode, it threw an undefined variable error. Explicitly test for EOF
and have it re-enter the goto prompt loop.
Signed-off-by: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
57886bc (git-add -i/-p: Change prompt separater from slash to comma,
2008-11-27) changed the prompt separator to ',', but forgot to adapt
the 'g' (goto) command.
Signed-off-by: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint:
User-manual: "git stash <comment>" form is long gone
add test-dump-cache-tree in Makefile
fix typo in Documentation
apply: fix access to an uninitialized mode variable, found by valgrind
* maint-1.6.0:
User-manual: "git stash <comment>" form is long gone
add test-dump-cache-tree in Makefile
fix typo in Documentation
apply: fix access to an uninitialized mode variable, found by valgrind
http-push: wrap signature of get_remote_object_url
The signature of get_remote_object_url stands at 96 characters (as
pointed out by Dscho); this patch wraps it so that it conforms to the
80 characters guideline.
5c5ba73 (Makefile: Use generic rule to build test programs,
2007-05-31) tried to use generic rule to build test programs, but it
misses the file 'dump-cache-tree.c', since its name is not prefixed by
'test-'. This commit solves this little problem by renaming this file
instead of carrying out an explicit rule in Makefile.
Signed-off-by: Guanqun Lu <guanqun.lu@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a bit of code to __git_ps1 that lets it append '*' to the branch
name if there are any unstaged changes, and '+' if there are any
staged changes.
Since this is a rather expensive operation and will force a lot of
data into the cache whenever you first enter a repository, you have to
enable it manually by setting GIT_PS1_SHOWDIRTYSTATE to a nonempty
value. The configuration variable bash.showDirtyState can then be
used to disable it again for some repositories.
Signed-off-by: Thomas Rast <trast@student.ethz.ch> Acked-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
apply: fix access to an uninitialized mode variable, found by valgrind
When 'tpatch' was initialized successfully, st_mode was already taken
from the previous diff. We should not try to override it with data
from an lstat() that was never called.
This is a companion patch to 7a07841(git-apply: handle a patch that
touches the same path more than once better).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
filter-branch: Fix fatal error on bare repositories
When git filter-branch is run on a bare repository, it prints out a fatal
error message:
$ git filter-branch branch
Rewrite 476c4839280c219c2317376b661d9d95c1727fc3 (9/9)
WARNING: Ref 'refs/heads/branch' is unchanged
fatal: This operation must be run in a work tree
Note that this fatal error message doesn't prevent git filter-branch from
exiting successfully. (Why doesn't git filter-branch actually exit with an
error when a shell command fails? I'm not sure why it was designed this
way.)
This error message is caused by the following section of code at the end of
git-filter-branch.sh:
if [ "$(is_bare_repository)" = false ]; then
unset GIT_DIR GIT_WORK_TREE GIT_INDEX_FILE
test -z "$ORIG_GIT_DIR" || {
GIT_DIR="$ORIG_GIT_DIR" && export GIT_DIR
}
... elided ...
git read-tree -u -m HEAD
fi
The problem is the call to $(is_bare_repository), which is made before
GIT_DIR and GIT_WORK_TREE are restored. This call always returns "false",
even when we're running in a bare repository. But this means that we will
attempt to call 'git read-tree' even in a bare repository, which will fail
and print an error.
This patch modifies git-filter-branch.sh to restore the original
environment variables before trying to call is_bare_repository.
tests: fix test_commit() for case insensitive filesystems
Brian Gernhardt noticed that t3411 was broken recently on case insensitive
filesystems.
0088496 (test-lib.sh: introduce test_commit() and test_merge() helpers,
2009-01-27) used a tag and a file with the same name, only different in
case, and converted many existing tests that needed only a file (or a
tag).
Some tests may want to refer to a rev or a file, but on a filesystem that
loses cases, referring to either without disambiguation mark ("--") on the
command line now triggers an error (t3411 was the only one such test).
Fix it by using a filename that is different from the tagname each step
creates.
2182896 (t3412: clean up GIT_EDITOR usage, 2009-01-30) tried to clean up
the script's use of GIT_EDITOR, but it can further be simplified, because
that is how test-lib.sh sets things up already.
receive-pack: explain what to do when push updates the current branch
This makes "git push" issue a more detailed instruction when a user pushes
into the current branch of a non-bare repository without having an
explicit configuration set to receive.denycurrentbranch. In such a case,
it will also tell the user that the default will change to refusal in a
future version of git.